Content identification method, device, medium and electronic device

By forming content recognition pairs and identifying them based on multiple dimension features, the misjudgment and misjudgment problems in content review are solved, automatic review is realized, and the accuracy and efficiency of identification results are improved.

CN113971402BActive Publication Date: 2024-12-06DOUYIN VISION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111235927.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-22
Publication Date
2024-12-06
Estimated Expiration
2041-10-22

AI Technical Summary

Technical Problem

The existing technology has problems of misjudgment and misjudgment in content review, and manual review consumes a lot of manpower, making it difficult to deal with the review process of large amounts of content.

Method used

By receiving the content to be identified, the content recognition pair is formed, and the content recognition pair is identified based on multiple dimension features. The content recognition model is used to determine the target recognition result of the content to be identified, so as to realize automatic review.

Benefits of technology

It improves the accuracy of identification results, reduces misjudgment and misjudgment, saves manual workload, and ensures that the release of original content provides effective data support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113971402B_ABST
    Figure CN113971402B_ABST
Patent Text Reader

Abstract

The present disclosure relates to a content recognition method, device, medium and electronic device, the method comprising: receiving content to be recognized; obtaining recognition results of the multiple content recognition pairs based on multiple content recognition pairs and content recognition models containing the content to be recognized, wherein each of the content recognition pairs also includes a candidate content in a preset set, and the content recognition model is used to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition results based on the multiple dimensional features, the multiple dimensions are used to represent multiple types of components in the content to be recognized; and determining the target recognition results of the content to be recognized based on the multiple recognition results. In this way, recognition between features of different dimensions can be achieved, the accuracy of the recognition results can be improved, manual workload can be saved, and it is convenient for users to use.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular, to a content identification method, device, medium and electronic device. Background Art

[0002] The development of computer technology has led to more and more attention to news, information and other content. At the same time, the rise of content sharing platforms has made the users who can publish content more diverse. Platforms usually carry a lot of UGC (User-generated content) and PGC (Professional-generated content). In order to ensure the validity of the content, staff are usually required to review the published content to determine whether the content to be published is plagiarized from other published content.

[0003] However, the above process requires a lot of manpower and is very likely to cause misjudgment or omission due to different subjective understandings of humans, and it is even more difficult to cope with the review process of a large amount of content. Summary of the invention

[0004] This summary is provided to introduce concepts in a brief form that will be described in detail in the detailed description below. This summary is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0005] In a first aspect, the present disclosure provides a content identification method, the method comprising:

[0006] Receiving content to be identified;

[0007] According to a plurality of content recognition pairs including the content to be recognized and a content recognition model, obtaining recognition results of the plurality of content recognition pairs, wherein each of the content recognition pairs further includes a candidate content in a preset set, and the content recognition model is used to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, wherein the multiple dimensions are used to represent multiple types of components in the content to be recognized;

[0008] According to the multiple recognition results, a target recognition result of the content to be recognized is determined.

[0009] In a second aspect, the present disclosure provides a content identification device, the device comprising:

[0010] A receiving module, used for receiving content to be identified;

[0011] a processing module, configured to obtain recognition results of the plurality of content recognition pairs according to the plurality of content recognition pairs including the content to be recognized and a content recognition model, wherein each of the content recognition pairs further includes a candidate content in a preset set, and the content recognition model is configured to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, wherein the multiple dimensions are used to represent multiple types of components in the content to be recognized;

[0012] The first determination module is used to determine a target recognition result of the content to be recognized based on the multiple recognition results.

[0013] In a third aspect, the present disclosure provides a computer-readable medium having a computer program stored thereon, which, when executed by a processing device, implements the steps of the method described in the first aspect.

[0014] In a fourth aspect, the present disclosure provides an electronic device, including:

[0015] a storage device having a computer program stored thereon;

[0016] A processing device is used to execute the computer program in the storage device to implement the steps of the method described in the first aspect.

[0017] In the above technical solution, by forming a content recognition pair between the content to be recognized and the existing content, the degree of association between the content to be recognized and the existing content is determined based on the content recognition pair, so that the recognition result of the content to be recognized can be comprehensively determined based on the results of multiple content recognition pairs, and the automatic review of the content to be recognized can be realized. Therefore, through the above technical solution, when the content to be recognized is recognized and reviewed, it can be recognized based on different types of components in the content to be recognized, so that similar overlapping behaviors between different components can be recognized, and recognition between features of different dimensions can be realized, thereby improving the accuracy of the recognition result, effectively avoiding the misjudgment and omission of the content to be recognized, effectively saving manual workload, and improving the accuracy of the recognition result. At the same time, it provides effective data support to ensure the release of original content, which is convenient for users to use.

[0018] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale. In the drawings:

[0020] Figure 1 is a flow chart of a content identification method provided according to an embodiment of the present disclosure;

[0021] Figure 2 is a flowchart of an exemplary implementation method for obtaining recognition results of multiple content recognition pairs based on multiple content recognition pairs including content to be recognized and a content recognition model;

[0022] Figure 3 is a schematic diagram of a title feature extraction sub-model provided according to an embodiment of the present disclosure;

[0023] Figure 4 is a schematic diagram of an image feature extraction sub-model provided according to an embodiment of the present disclosure;

[0024] Figure 5 is a structural diagram of a content identification model provided according to an embodiment of the present disclosure;

[0025] Figure 6 is a block diagram of a content recognition device provided according to an embodiment of the present disclosure;

[0026] Figure 7 A schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0027] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0028] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0029] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0030] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0031] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0032] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0033] In actual application scenarios, the applicant has found through research that the following plagiarism phenomena may occur when writing content: copying images in published content, that is, copying the text in the image in the existing content in the form of text to form new text content; or modifying the video subtitles in the published content into text, that is, extracting the subtitle information in the video in the existing content and copying it to form new text content; or simply rewriting another content to form new content. In the above process, there may be a situation where the similarity between different types of information is too high, and the related technology may have the problem of missed judgment for such phenomenon. However, the corresponding contents of articles with similar titles may be quite different, and it is easy to make misjudgments for such phenomenon. It is difficult to perform effective identification in the prior art, and the workload required for manual identification increases dramatically. Based on this, the present disclosure also provides the following embodiments.

[0034] Figure 1 As shown, it is a flowchart of a content identification method provided according to an embodiment of the present disclosure, as shown Figure 1 As shown, the method may include:

[0035] In step 11, the content to be identified is received, wherein the content to be identified may be uploaded content that needs to be reviewed, such as uploaded UGC content, PGC content, or content uploaded by an official account, which may be determined based on actual needs for review and identification.

[0036] In step 12, recognition results of multiple content recognition pairs are obtained based on multiple content recognition pairs containing content to be recognized and content recognition models, wherein each of the content recognition pairs also includes a candidate content in a preset set, and the content recognition model is used to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, and the multiple dimensions are used to represent various types of components in the content to be recognized, such as dimensions that may be text, title, image, audio, video, etc., so that recognition can be performed based on features in multiple dimensions, thereby achieving recognition between features of different dimensions and improving the accuracy of the recognition result.

[0037] Among them, the preset collection can be set according to the actual usage scenario. For example, the preset collection can be a collection formed by all content data that can be obtained by the platform for review, or it can be a collection formed for a certain type of content data, such as a sports news collection, an entertainment news collection, etc. The present disclosure does not limit this.

[0038] In a possible embodiment, the candidate content may be determined in the following manner:

[0039] The similarity between the content to be identified and the plurality of stored contents in the preset set is calculated, so as to obtain the similarity between the content to be identified and the stored contents. For example, the content to be identified and the stored contents can be converted into their respective vector representations based on a consistent vector conversion method, and then similarity calculation can be performed based on the vector representations, such as calculating the inverse of the distance, the cosine value of the angle, etc.

[0040] Afterwards, candidate content is determined based on the similarity between the content to be identified and the stored content. For example, the stored content with a similarity greater than a threshold can be determined as candidate content, or the stored content can be sorted in descending order of similarity to obtain the top N stored content as candidate content. N can be set according to the actual application scenario, such as 10 or 15, to minimize the amount of data calculation while ensuring calculation accuracy.

[0041] Therefore, the determined candidate content is content that has a certain degree of similarity with the content to be identified. In this step, the content to be identified and any candidate content can form a content identification pair to further determine whether the content similarity between the two is too high through a one-to-one comparison.

[0042] In step 13, a target recognition result of the content to be recognized is determined based on the multiple recognition results, wherein the target recognition result is used to indicate the originality of the content to be recognized.

[0043] In the above technical solution, by forming a content recognition pair between the content to be recognized and the existing content, the degree of association between the content to be recognized and the existing content is determined based on the content recognition pair, so that the recognition result of the content to be recognized can be comprehensively determined based on the results of multiple content recognition pairs, and the automatic review of the content to be recognized can be realized. Therefore, through the above technical solution, when the content to be recognized is recognized and reviewed, it can be recognized based on different types of components in the content to be recognized, so that similar overlapping behaviors between different components of the published content can be recognized, and the recognition between features of different dimensions can be realized, thereby improving the accuracy of the recognition result, effectively avoiding the misjudgment and omission of the content to be recognized, effectively saving manual workload, and improving the accuracy of the recognition result. At the same time, it provides effective data support to ensure the release of original content, which is convenient for users to use.

[0044] In a possible embodiment, the dimension corresponds one-to-one to the feature extraction sub-model of the content recognition model. For example, the dimension may be title, text, image, video, audio, etc., and the content recognition model may include a title feature extraction sub-model, a text feature extraction sub-model, an image feature extraction sub-model, a video feature extraction sub-model, and an audio feature extraction sub-model.

[0045] Accordingly, in step 12, an exemplary implementation method of obtaining recognition results of multiple content recognition pairs based on multiple content recognition pairs including the content to be recognized and the content recognition model is as follows: Figure 2 As shown, this step may include:

[0046] In step 21, for each content identification pair, dimensional features corresponding to the content to be identified and the candidate content in the content identification pair, as well as the feature extraction sub-model corresponding to each dimension, are determined respectively in multiple dimensions for the content to be identified and the candidate content.

[0047] The structures of different feature extraction sub-models may be the same or different, and the present disclosure does not limit this. For example, the title feature extraction sub-model may be to segment the title, and perform Word Embedding based on the segmentation, such as obtaining the word vector of the segmentation through word2vec, and then generating the Embedding vector corresponding to the title based on the word vector. Therefore, the title of the content to be identified is input into the title feature extraction sub-model to obtain the Embedding vector corresponding to the content to be identified in the title dimension, such as Figure 3 As shown in area A in the figure; similarly, the title of the candidate content is input into the title feature extraction sub-model to obtain the Embedding vector corresponding to the content to be identified in the title dimension, as shown in Figure 3 Shown in area B.

[0048] As an example, the text feature extraction sub-model can have the same structure as the title feature extraction sub-model, that is, the text feature extraction sub-model performs word segmentation, performs word embedding word segmentation based on the word segmentation, and then generates the embedding vector corresponding to the text based on the word vector.

[0049] For example, the image feature extraction sub-model can be a CNN (Convolutional Neural Networks) or ResNet residual network to extract features, thereby obtaining the Embedding vector corresponding to the image, that is, the image embedding vector. Among them, the content to be identified and the candidate content can contain multiple images, such as Figure 4 As shown in the figure, the images in the content to be identified are represented as D1-Dn, and the images in the candidate content are represented as H1-Hm. Then, the features of each image can be extracted in turn according to the display order of the images and the Image Embedding vector can be encoded to obtain the Embedding vector corresponding to the content to be identified in the image dimension, as shown in Figure 4 shown.

[0050] Among them, for the audio dimension, the audio feature extraction submodel can recognize the audio therein by means of speech recognition, obtain the audio corresponding to the text, and then determine the vector corresponding to the recognized text as the vector corresponding to the audio. For the video dimension, the video feature extraction submodel can recognize the audio therein by automatic speech recognition (ASR, Automatic Speech Recognition) technology, or recognize the subtitles in the video by optical character recognition (OCR, Optical Character Recognition) to obtain the video corresponding to the text, and obtain the image in the video at intervals of a preset time period to obtain an image sequence, so that the vector corresponding to the text and the vector corresponding to each image in the image sequence can be determined based on the method described above to obtain the vector corresponding to the video.

[0051] In step 22, for each dimension, the features of the content to be identified and the candidate content in the dimension are concatenated to obtain the concatenated features corresponding to the dimension.

[0052] For example, Figure 3As shown, area A is used to obtain dimensional features of the content to be identified, and area B is used to obtain dimensional features of the candidate content. In the embodiment of the present disclosure, the content to be identified is not directly compared with the candidate content, but the content to be identified and the candidate content are fused together to obtain a feature for identification based on the feature. Therefore, when splicing the features of the content to be identified and the candidate content, a separation mark can be used to separate the features of the content to be identified and the candidate content, so that the features of the two can be effectively distinguished while the features of the two are spliced. Among them, the setting of the separation mark can be set according to the specific application scenario, and vector encoding can be performed on it, so that it can be marked as a separation vector in the splicing feature, such as Figure 3 As shown in area C. Similarly, Figure 4 As shown, in the image dimension, the segmentation mark can be a segmentation map, so that the feature extraction and Embedding vector encoding of the segmentation map can be performed to obtain the Embedding vector corresponding to the segmentation map, and then the splicing feature can be obtained.

[0053] In step 23, based on the concatenated features corresponding to the multiple dimensions, the fused features corresponding to the content identification pair are obtained.

[0054] In step 24, based on the classification sub-model and the fusion features of the content recognition model, the recognition result of the content recognition pair is determined.

[0055] As an example, the splicing features corresponding to the multiple dimensions can be spliced ​​based on the direction of dimension splicing to obtain the fused feature, that is, the fused feature contains the splicing features corresponding to the multiple dimensions. Furthermore, the fused feature contains the features of the content to be identified and the candidate content at the same time, so that identification and classification can be performed based on the fused feature and the classification sub-model, so that the classification result can be obtained.

[0056] Therefore, through the above technical solution, feature extraction can be performed for each dimension of the candidate content and the content to be identified, so that the features of the content to be identified and the candidate content in each dimension can be obtained, and then the features in multiple dimensions can be fused, so that the fused features can simultaneously contain the features of the content to be identified and the candidate content in multiple dimensions, ensuring the accuracy and comprehensiveness of the features for classification and identification, and providing accurate data support for the classification of the identification results. At the same time, classification can be performed based on the fused features, and compared with the calculation of similarity in the related technology, more classification results can be obtained, so as to expand the content identification method and improve the application scope of the content identification method.

[0057] In a possible embodiment, in step 23, an exemplary implementation method of obtaining fusion features corresponding to the content identification pair according to the splicing features corresponding to the multiple dimensions is as follows, and this step may include:

[0058] For the splicing features corresponding to each dimension, the splicing features are processed based on the first attention layer to obtain the attention features corresponding to the splicing features.

[0059] For example, the splicing features can be processed based on the Transformer model, wherein the Transformer model includes an attention layer, so that when processing the splicing features, the attention information of the content to be identified on the candidate content and the self-attention information of the content to be identified can be fused on the basis of the splicing features, so as to facilitate the subsequent determination of similar parts in the candidate content and the content quality information of the content to be identified.

[0060] As an example, in order to further improve the accuracy of feature processing, a multi-layer Transformer model can be used for processing. For example, for the main text and image parts with complex content information, a double-layer Transformer can be used for attention processing, and for the title part, a single-layer Transformer can be used for attention processing.

[0061] The attention features in each dimension are spliced ​​together to obtain multi-dimensional splicing features.

[0062] For example, the size of the attention features in each dimension is the same. For example, if the attention features in each dimension are a 10*32 matrix, the attention features can be spliced ​​based on the dimension splicing method. After splicing the attention features of three dimensions such as title, text, and image, a multi-dimensional spliced ​​feature with a size of 10*96 can be obtained.

[0063] The multi-dimensional splicing features are processed based on the second attention layer to obtain the fusion features.

[0064] Among them, the structure of the second attention layer and the first attention layer can be the same or different, and it can be set according to the actual application scenario. The attention layer processes the multi-dimensional splicing features again, so that the multi-dimensional splicing features can be included in the fusion features, and different weights can be given to the information in the multi-dimensional splicing features. Among them, through attention processing, the correlation between them can be found based on the original data and the important features can be highlighted. For example, the multi-dimensional splicing features can include the title features, text features and image features of the content to be identified, as well as the title features, text features and image features of the candidate content. Therefore, through the attention calculation processing of the second attention layer, the title features in the content to be identified can also pay attention to the title features, text features and image features of the candidate content, so as to obtain the association between the features in the content to be identified and the features under multiple dimensions in the candidate content, thereby ensuring the accuracy of the subsequent identification of the content to be identified, and providing data support for the identification of the similarity between different types of information to further determine the identification of plagiarism or duplication.

[0065] In a possible embodiment, the recognition result includes a recognition parameter of the content to be recognized corresponding to a classification, wherein the classification is used to indicate the originality of the content to be recognized, and the recognition parameter may be a probability that the content to be recognized corresponds to the classification.

[0066] Accordingly, an exemplary implementation of determining the recognition result of the content recognition pair based on the classification sub-model of the content recognition model and the fusion feature may include:

[0067] According to the fusion feature and the classification sub-model, feature vectors of the fusion feature corresponding to multiple categories are obtained, wherein the categories include similar content category, low-quality content category, and original content category, wherein the dimensions in the feature vector correspond to the categories in the multiple categories one by one.

[0068] Among them, the similar content classification is used to indicate that the overlap between the content to be identified and the existing content is too high, that is, the content to be identified is similar content; the low-quality content classification is used to indicate that the content to be identified is difficult to identify and review due to problems such as the small number of words in the content to be identified or the blurred image or unclear meaning; the original content classification is used to indicate that the overlap between the content to be identified and the existing content is low, that is, the content to be identified is original content.

[0069] Among them, the classification sub-model can be implemented based on the GAP (Global Average Pooling) layer and the Dense dense connection layer. For example, the fused features can be input into the GAP layer to globally average the feature map and output it to the Dense layer to obtain a feature vector for classification. The GAP layer can make the conversion between the input fused features and the determined feature vector for classification simpler, and compared with the FC fully connected layer, this layer does not require a large number of training and tuning parameters, so that the spatial parameters can be effectively reduced to make the model more robust and avoid model overfitting. Among them, the structure of the DAP layer and the Dense layer can select the conventional structure in the field, and the present disclosure does not limit this.

[0070] The feature vector is processed to obtain the recognition parameter of the fused feature corresponding to each category, so as to obtain the recognition result.

[0071] Afterwards, the feature vector can be processed by a softmax activation function to determine the probability that the feature vector corresponds to each category based on the features of each dimension in the feature vector, that is, to obtain the recognition parameters of each category.

[0072] Through the above technical scheme, when identifying the content to be identified, compared with the scheme in the related art that calculates the similarity between the content to be identified and the existing content to determine whether it is similar to the published content, the present disclosure combines the two contents in the content identification pair into a fusion feature for comprehensive judgment, thereby effectively identifying the phenomenon of copying and plagiarism between different types of content, such as "copying images in published content" and "modifying video subtitles in published content into text", so as to effectively identify the plagiarism or copying behaviors between different types of information, thereby providing effective data support for the subsequent determination of the accuracy of the recognition results of the content to be identified.

[0073] In a possible embodiment, the recognition result includes a recognition parameter of the content to be recognized corresponding to a classification, wherein the classification is used to indicate the originality of the content to be recognized, and the recognition parameter may be a probability that the content to be recognized corresponds to the classification.

[0074] Accordingly, an exemplary implementation of determining a target recognition result of the content to be recognized based on the multiple recognition results is as follows. This step may include:

[0075] Recognition parameters corresponding to similar content categories in the plurality of recognition results are obtained, that is, probability values ​​corresponding to similar content categories determined in the plurality of content recognition pairs are determined.

[0076] If the identification parameters of the plurality of similar content categories satisfy the similar identification condition, the target identification result is determined to be a similar content category, wherein the similar identification condition may include the following situations:

[0077] In a first case, the maximum value of the identification parameter corresponding to the similar content classification is greater than a first preset threshold.

[0078] Among them, as described above, multiple candidate contents can be preliminarily screened out based on the content to be identified, and then the content to be identified is compared with each candidate identification content one by one. As an example, if it is determined that the probability value of the content to be identified corresponding to a similar content classification compared to a candidate content is too large, it can be determined that the similarity between the content to be identified and the candidate content is too high, and at this time, the target recognition result of the content to be identified can be determined to be a similar content classification. Furthermore, in the present disclosure, judgment can be made directly based on the maximum value of the recognition parameter corresponding to the similar content classification, thereby reducing the amount of data required for comparison and judgment, ensuring the accuracy of the target recognition result while improving the recognition efficiency.

[0079] As another example, if the maximum value of the recognition parameter corresponding to the similar content classification is less than or equal to the first preset threshold, multiple recognition results corresponding to the content to be recognized can be output for manual review to further determine the target recognition result of the content to be recognized.

[0080] In the second case, the maximum value of the recognition parameters corresponding to the similar content classification is less than or equal to the first preset threshold, and the average value of the recognition parameters corresponding to the similar content classification is greater than the second preset threshold. At this time, it can also be determined that the target recognition result is a similar content classification.

[0081] Among them, if the maximum value of the identification parameter corresponding to the similar content classification is less than or equal to the first preset threshold, it means that the probability of the content to be identified and any candidate content corresponding to the similar content classification does not meet the requirement, that is, the similarity between the content to be identified and any candidate content is not high, and the content to be identified is not highly similar to any candidate content. In this case, it can be further determined whether the content to be identified is similar to multiple candidate contents at the same time.

[0082] Accordingly, in the embodiment of the present disclosure, when the maximum value of the identification parameter corresponding to the similar content classification is less than or equal to the first preset threshold, the average value of the identification parameter corresponding to the similar content classification can be further determined. When the average value of the identification parameter corresponding to the similar content classification is greater than the second preset threshold, it means that the content to be identified is more similar to each candidate content, that is, the content to be identified is similar to multiple candidate contents, which means that the content to be identified may be a partial content copy selected from multiple candidate contents at the same time. At this time, the target recognition result of the content to be identified can be determined as a similar content classification.

[0083] Therefore, through the above-mentioned technical scheme, the recognition results of multiple content recognition pairs can be combined for comprehensive determination, which can not only determine the similarity between the content to be recognized and a single content, but also determine the similarity between the content to be recognized and multiple contents. In this way, a comprehensive analysis of the content to be recognized can be carried out, and the comprehensiveness and accuracy of the recognition results of the content to be recognized can be improved, while the scope of use of the disclosed scheme can be effectively expanded.

[0084] In another possible embodiment, an exemplary implementation of determining a target recognition result of the content to be recognized based on the multiple recognition results is as follows. This step may also include:

[0085] Recognition parameters corresponding to low-quality content categories in the plurality of recognition results are obtained, that is, probability values ​​corresponding to low-quality content categories determined in the plurality of content recognition pairs are determined.

[0086] If the identification parameters of the plurality of low-quality content classifications satisfy the low-quality identification condition, and the identification parameters of the plurality of similar content classifications do not satisfy the similar identification condition, it is determined that the target identification result is a low-quality content classification, wherein the low-quality identification condition may include the following situations:

[0087] In a first case, the maximum value of the identification parameter corresponding to the low-quality content classification is greater than a third preset threshold.

[0088] Among them, whether the content to be identified is low-quality content can be determined based on the identification score corresponding to the low-quality content classification in the identification results of each content identification pair. For example, if it is determined that the probability value of the content to be identified corresponding to the low-quality content classification is too large, then the content to be identified may be low-quality content. If the content to be identified does not belong to the similar content classification, then it can be determined that the target identification result of the content to be identified is a low-quality content classification. Furthermore, in the present disclosure, judgment can be made directly based on the maximum value of the identification parameter corresponding to the low-quality content classification, thereby reducing the amount of data required for comparison and judgment, ensuring the accuracy of the target identification result while improving the identification efficiency.

[0089] In the second case, the maximum value of the identification parameter corresponding to the low-quality content classification is less than or equal to the third preset threshold, and the average value of the identification parameter corresponding to the low-quality content classification is greater than the fourth preset threshold. At this time, it can also be determined that the target recognition result is a low-quality content classification.

[0090] Among them, the maximum value of the identification parameter corresponding to the low-quality content classification is less than or equal to the third preset threshold, indicating that there is no clear directional result in the multiple identification results of the content to be identified. Therefore, the average value of the identification parameter corresponding to the low-quality content classification can be further determined. When the average value of the identification parameter corresponding to the low-quality content classification is greater than the fourth preset threshold, it means that the probability that the multiple identification results of the content to be identified comprehensively point to it as low-quality content is relatively high. At this time, the target identification result of the content to be identified can be determined as a low-quality content classification.

[0091] Therefore, through the above-mentioned technical scheme, the recognition results of multiple content recognition pairs can be combined for comprehensive determination, so that it is possible to review whether the content to be recognized is highly similar to the published content, and at the same time, the quality of the content to be recognized itself can be reviewed, so that in the process of content recognition, we can directly focus on the writing quality of the content to be recognized itself, avoid one-sided review caused by directly determining the content to be recognized as similar content or original content, improve the accuracy of content recognition, and thus improve the comprehensiveness and accuracy of the recognition results of the content to be recognized, which can further expand the scope of use of the disclosed scheme, and at the same time provide data support for avoiding the release of low-quality content and improving user experience.

[0092] In another possible embodiment, the exemplary implementation method of determining a target recognition result of the content to be recognized according to the multiple recognition results may also include:

[0093] Recognition parameters corresponding to the original content categories in the plurality of recognition results are obtained, that is, probability values ​​corresponding to the original content categories determined in the plurality of content recognition pairs are determined.

[0094] If the maximum value of the identification parameters of the multiple original content classifications is greater than the fifth preset threshold, and the identification parameters of the multiple similar content classifications do not meet the similarity identification condition, and the identification parameters of the multiple low-quality content classifications do not meet the low-quality identification condition, it is determined that the target identification result is an original content classification.

[0095] In this embodiment, when it is determined that the maximum value of the identification parameter corresponding to the original content classification of the content to be identified compared with a certain candidate content is greater than the fifth preset threshold, it means that the content to be identified is original content compared with the candidate content. If the content to be identified is neither similar content nor low-quality content compared with other candidate contents, the content to be identified can be directly determined as original content.

[0096] It should be noted that the settings of the first preset threshold, the second preset threshold, the third preset threshold, the fourth preset threshold, and the fifth preset threshold can be set according to actual application scenarios, and the present disclosure does not limit this.

[0097] Therefore, through the above technical solution, when the content to be identified is classified as original content compared to any candidate content, it cannot be directly determined as original content. It is necessary to further determine that the content to be identified does not have too high similarity or low-quality writing with other candidate content before it can be determined that the content to be identified is original content. This can effectively ensure the accuracy of the identification results of the content to be identified and provide accurate data support for the subsequent release of content.

[0098] In another possible embodiment, the maximum and average values ​​of the identification parameters corresponding to the original content classification, low-quality content classification and similar content classification in the multiple identification results can be displayed and output, so that the user can make an audit decision directly based on the information of each identification parameter during the audit, thereby further ensuring the accuracy of the target recognition results of the content to be identified, reducing the manual workload to a certain extent, and reducing the misjudgment and missed judgment of content with too high similarity, thereby improving the efficiency and accuracy of content recognition.

[0099] In the case where the content to be identified is determined to be similar content, in order to further illustrate the accuracy of the identification result, the similar part with a high similarity corresponding to the content to be identified can be determined, that is, the candidate content in the content identification pair with the maximum value of the identification parameter can be directly determined as the similar part, and the content identification pair is determined as the target content identification pair. As an example, the candidate content in the target content identification pair can be directly displayed and output to show the user the similar part corresponding to the content to be identified.

[0100] In another possible embodiment, the method may further include:

[0101] When it is determined that the target recognition result is a similar content classification, the target position whose attention parameter in the fusion feature is greater than a preset threshold is determined according to the fusion feature corresponding to the target content recognition pair, wherein the target content recognition pair is a content recognition pair corresponding to the maximum value of the recognition parameter of the similar content classification.

[0102] Among them, in the embodiment of the present disclosure, classification is performed based on the fusion feature and the classification sub-model to obtain the recognition result. Therefore, when the content to be recognized is determined to be similar content based on the fusion feature, the part with a larger attention parameter in the recognition and classification process has a greater impact on the recognition result, that is, the target position in the fusion feature with an attention parameter greater than a preset threshold contributes more to the classification of the content to be recognized as similar content, and the content corresponding to the target position can be regarded as the similar part.

[0103] Further, the content corresponding to the target position in the candidate content of the target content recognition pair is determined as the overlapping content corresponding to the content to be recognized;

[0104] The candidate content and the overlapping content in the target content identification pair are output.

[0105] When extracting features from the candidate content, the extraction is performed based on the order of the parts of the candidate content. For example, the extraction is performed sequentially according to each paragraph of the main text, and the main text features A1-Am correspond to the features of the first to mth paragraphs in the main text, respectively, so that the part of the content corresponding to the target position in the candidate content can be determined directly based on the determined target position, that is, the overlapping part of the content with a high similarity to the content to be identified. Then, the candidate content and the overlapping content are displayed.

[0106] Therefore, through the above technical solution, the user can be prompted that the content to be identified is similar content, and the user can be further prompted with existing content with a high similarity to the content to be identified, as well as the similar parts in the existing content, so as to provide data support for ensuring the accuracy of the target recognition results, provide a clearer explanation and display of the target recognition results, improve the credibility of the target recognition results, and enhance the user experience.

[0107] In a possible embodiment, the dimensions correspond one-to-one to the feature extraction sub-models of the content recognition model. For example, the dimensions include title, text, and image. Figure 5 The above is a schematic diagram of the structure of the content identification model, and the content identification model is obtained in the following manner:

[0108] The training sample data is obtained, wherein the training sample data includes sample content, associated content corresponding to the sample content, and associated tags. The associated content may be similar content determined to be corresponding to the sample content, and the associated tag may indicate the originality of the sample content relative to the associated content, such as the low-quality content classification, original content classification, and similar content classification described above.

[0109] For each of the training sample data, according to the sample content and the associated content corresponding to the sample content, and the feature extraction sub-model corresponding to each dimension, determine the dimensional features corresponding to the sample content and the associated content in multiple dimensions respectively;

[0110] For each dimension, the features of the sample content and the related content under the dimension are spliced ​​to obtain the spliced ​​features corresponding to the dimension; based on the spliced ​​features corresponding to multiple dimensions, the fused features corresponding to the training sample data are obtained; based on the classification sub-model of the content recognition model and the fused features, the recognition results of the training sample data are determined.

[0111] The specific implementation of the above steps is the same as the related processing flow between the content to be identified and the candidate content described above, and will not be repeated here.

[0112] According to the recognition result and the associated label, the parameters of the content recognition model are adjusted to obtain a trained content recognition model.

[0113] Among them, the recognition error of the content recognition model can be calculated based on the recognition result and the associated label. When the recognition error is greater than an error threshold, the parameters of the content recognition model are adjusted based on the recognition error. For example, the parameters can be adjusted based on the gradient descent method. The training is stopped until the calculated recognition error of the recognition model is less than or equal to the error threshold, thereby obtaining a trained content recognition model.

[0114] Therefore, through the above technical solution, the content recognition model can be trained based on the sample content, the related content corresponding to the sample content, and the related labels, so as to realize the feature extraction, splicing and fusion, and classification recognition of the two contents used for comparison. By aggregating and judging the features under different dimensions, the content recognition model can be used to identify different types of content, thereby improving the application scope of the content recognition model and improving the accuracy of the recognition results obtained based on the content recognition model.

[0115] The present disclosure also provides a content recognition device, such as Figure 6 As shown, the device 10 comprises:

[0116] The receiving module 100 is used to receive content to be identified;

[0117] The processing module 200 is used to obtain recognition results of the multiple content recognition pairs according to the multiple content recognition pairs including the content to be recognized and the content recognition model, wherein each of the content recognition pairs also includes a candidate content in a preset set, and the content recognition model is used to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, and the multiple dimensions are used to represent multiple types of components in the content to be recognized;

[0118] The first determination module 300 is used to determine a target recognition result of the content to be recognized according to the multiple recognition results.

[0119] Optionally, the dimensions correspond one-to-one to the feature extraction sub-models of the content recognition model, and the processing module includes:

[0120] A first determination submodule is used to determine, for each content recognition pair, dimensional features corresponding to the content to be recognized and the candidate content in multiple dimensions according to the content to be recognized and the candidate content in the content recognition pair, and the feature extraction submodel corresponding to each dimension;

[0121] A first splicing submodule is used to splice the features of the content to be identified and the candidate content in each dimension to obtain a splicing feature corresponding to the dimension;

[0122] A fusion submodule, used for obtaining fusion features corresponding to the content recognition pair according to the splicing features corresponding to the multiple dimensions;

[0123] The second determination submodule is used to determine the recognition result of the content recognition pair based on the classification submodel of the content recognition model and the fusion feature.

[0124] Optionally, the fusion submodule includes:

[0125] A first processing submodule is used to process the splicing features corresponding to each dimension based on the first attention layer to obtain attention features corresponding to the splicing features;

[0126] The second splicing submodule is used to splice the attention features in each dimension to obtain multi-dimensional splicing features;

[0127] The second processing submodule is used to process the multi-dimensional splicing features based on the second attention layer to obtain the fusion features.

[0128] Optionally, the recognition result includes recognition parameters of the content to be recognized corresponding to a classification, and the second determination submodule includes:

[0129] A third processing submodule is used to obtain feature vectors of the fused features corresponding to a plurality of classifications according to the fused features and the classification submodel, wherein the classifications include similar content classification, low-quality content classification, and original content classification;

[0130] The fourth processing submodule is used to process the feature vector to obtain the recognition parameter of the fused feature corresponding to each category to obtain the recognition result.

[0131] Optionally, the recognition result includes recognition parameters corresponding to the classification of the content to be recognized, and the first determination module includes:

[0132] A first acquisition submodule is used to acquire recognition parameters corresponding to similar content categories in a plurality of recognition results;

[0133] The third determination submodule is configured to determine that the target recognition result is a similar content classification if the recognition parameters of the plurality of similar content classifications satisfy a similar recognition condition, wherein the similar recognition condition includes:

[0134] The maximum value of the identification parameter corresponding to the similar content classification is greater than a first preset threshold;

[0135] The maximum value of the identification parameters corresponding to the similar content classification is less than or equal to the first preset threshold, and the average value of the identification parameters corresponding to the similar content classification is greater than a second preset threshold.

[0136] Optionally, the first determining module further includes:

[0137] A second acquisition submodule is used to acquire recognition parameters corresponding to the low-quality content classifications in the plurality of recognition results;

[0138] The fourth determination submodule is configured to determine that the target recognition result is a low-quality content classification if the recognition parameters of the plurality of low-quality content classifications satisfy the low-quality recognition condition and the recognition parameters of the plurality of similar content classifications do not satisfy the similar recognition condition, wherein the low-quality recognition condition includes:

[0139] The maximum value of the identification parameter corresponding to the low-quality content classification is greater than a third preset threshold;

[0140] The maximum value of the identification parameter corresponding to the low-quality content classification is less than or equal to the third preset threshold, and the average value of the identification parameter corresponding to the low-quality content classification is greater than a fourth preset threshold.

[0141] Optionally, the first determining module further includes:

[0142] A third acquisition submodule is used to obtain recognition parameters corresponding to the original content categories in the multiple recognition results;

[0143] The fifth determination submodule is used to determine that the target recognition result is an original content classification if the maximum value of the recognition parameters of the multiple original content classifications is greater than the fifth preset threshold, and the recognition parameters of the multiple similar content classifications do not meet the similar recognition condition, and the recognition parameters of the multiple low-quality content classifications do not meet the low-quality recognition condition.

[0144] Optionally, the device further comprises:

[0145] A second determination module is used to determine, when it is determined that the target recognition result is a similar content classification, a target position whose attention parameter in the fusion feature is greater than a preset threshold value according to the fusion feature corresponding to the target content recognition pair, wherein the target content recognition pair is a content recognition pair corresponding to the maximum value of the recognition parameter of the similar content classification;

[0146] A third determination module is used to determine the content corresponding to the target position in the candidate content of the target content recognition pair as the overlapping content corresponding to the content to be recognized;

[0147] An output module is used to output the candidate content in the target content recognition pair and the overlapping content.

[0148] Optionally, the dimensions correspond one-to-one to the feature extraction sub-models of the content identification model, and the content identification model is obtained in the following manner:

[0149] Acquire the training sample data, wherein the training sample data includes sample content, associated content corresponding to the sample content, and associated tags;

[0150] For each of the training sample data, according to the sample content and the associated content corresponding to the sample content, and the feature extraction sub-model corresponding to each dimension, determine the dimensional features corresponding to the sample content and the associated content in multiple dimensions respectively;

[0151] For each dimension, concatenate the features of the sample content and the associated content under the dimension to obtain a concatenated feature corresponding to the dimension;

[0152] Obtaining fusion features corresponding to the training sample data according to the splicing features corresponding to the multiple dimensions;

[0153] Determining the recognition result of the training sample data based on the classification sub-model of the content recognition model and the fusion feature;

[0154] According to the recognition result and the associated label, the parameters of the content recognition model are adjusted to obtain a trained content recognition model.

[0155] Reference below Figure 7 , which shows a schematic diagram of the structure of an electronic device 600 suitable for implementing the embodiment of the present disclosure. The terminal device in the embodiment of the present disclosure may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0156] like Figure 7 As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0157] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 7 The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0158] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0159] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0160] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0161] The computer-readable medium may be included in the electronic device, or may exist independently without being installed in the electronic device.

[0162] The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed by the electronic device, the electronic device: receives content to be identified; obtains identification results of the multiple content identification pairs and content identification models containing the content to be identified, wherein each of the content identification pairs also includes a candidate content in a preset set, and the content identification model is used to obtain dimensional features of the content to be identified and the candidate content in multiple dimensions, and determines the identification result based on the multiple dimensional features, and the multiple dimensions are used to represent multiple types of components in the content to be identified; and determines the target identification result of the content to be identified based on the multiple identification results.

[0163] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0164] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0165] The modules involved in the embodiments described in the present disclosure may be implemented by software or hardware. The name of a module does not limit the module itself in some cases. For example, a receiving module may also be described as a "module for receiving content to be identified".

[0166] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0167] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0168] According to one or more embodiments of the present disclosure, Example 1 provides a content identification method, wherein the method includes:

[0169] Receiving content to be identified;

[0170] According to a plurality of content recognition pairs including the content to be recognized and a content recognition model, obtaining recognition results of the plurality of content recognition pairs, wherein each of the content recognition pairs further includes a candidate content in a preset set, and the content recognition model is used to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, wherein the multiple dimensions are used to represent multiple types of components in the content to be recognized;

[0171] According to the multiple recognition results, a target recognition result of the content to be recognized is determined.

[0172] According to one or more embodiments of the present disclosure, Example 2 provides the method of Example 1, wherein the dimension corresponds one-to-one to the feature extraction sub-model of the content recognition model, and obtaining recognition results of the multiple content recognition pairs based on the multiple content recognition pairs containing the content to be recognized and the content recognition model includes:

[0173] For each of the content identification pairs, determining dimensional features corresponding to the content to be identified and the candidate content in the content identification pair, respectively, in multiple dimensions according to the content to be identified and the candidate content, and the feature extraction sub-model corresponding to each dimension;

[0174] For each dimension, concatenate the features of the content to be identified and the candidate content under the dimension to obtain a concatenated feature corresponding to the dimension;

[0175] Obtaining fusion features corresponding to the content recognition pair according to the splicing features corresponding to the multiple dimensions;

[0176] Based on the classification sub-model of the content recognition model and the fusion feature, a recognition result of the content recognition pair is determined.

[0177] According to one or more embodiments of the present disclosure, Example 3 provides the method of Example 2, wherein obtaining the fusion feature corresponding to the content recognition pair according to the splicing features corresponding to the multiple dimensions includes:

[0178] For each dimension, the splicing feature is processed based on the first attention layer to obtain an attention feature corresponding to the splicing feature;

[0179] Concatenate the attention features in each dimension to obtain multi-dimensional concatenated features;

[0180] The multi-dimensional splicing features are processed based on the second attention layer to obtain the fusion features.

[0181] According to one or more embodiments of the present disclosure, Example 4 provides the method of Example 2, wherein the recognition result includes recognition parameters corresponding to the classification of the content to be recognized, and the determination of the recognition result of the content recognition pair based on the classification sub-model of the content recognition model and the fusion feature includes:

[0182] According to the fusion feature and the classification sub-model, feature vectors of the fusion feature corresponding to a plurality of classifications are obtained, wherein the classifications include similar content classification, low-quality content classification, and original content classification;

[0183] The feature vector is processed to obtain the recognition parameter of the fused feature corresponding to each category, so as to obtain the recognition result.

[0184] According to one or more embodiments of the present disclosure, Example 5 provides the method of Example 1, wherein the recognition result includes a recognition parameter of the content to be recognized corresponding to a classification, and determining a target recognition result of the content to be recognized based on the plurality of recognition results includes:

[0185] Obtaining recognition parameters corresponding to similar content categories in multiple recognition results;

[0186] If the identification parameters of the plurality of similar content classifications satisfy similar identification conditions, the target identification result is determined to be a similar content classification, wherein the similar identification conditions include:

[0187] The maximum value of the identification parameter corresponding to the similar content classification is greater than a first preset threshold;

[0188] The maximum value of the identification parameters corresponding to the similar content classification is less than or equal to the first preset threshold, and the average value of the identification parameters corresponding to the similar content classification is greater than a second preset threshold.

[0189] According to one or more embodiments of the present disclosure, Example 6 provides the method of Example 5, wherein the step of determining a target recognition result of the content to be recognized based on the plurality of recognition results further includes:

[0190] Obtaining recognition parameters corresponding to low-quality content categories respectively in the plurality of recognition results;

[0191] If the identification parameters of the plurality of low-quality content classifications satisfy the low-quality identification condition, and the identification parameters of the plurality of similar content classifications do not satisfy the similar identification condition, it is determined that the target identification result is a low-quality content classification, wherein the low-quality identification condition includes:

[0192] The maximum value of the identification parameter corresponding to the low-quality content classification is greater than a third preset threshold;

[0193] The maximum value of the identification parameter corresponding to the low-quality content classification is less than or equal to the third preset threshold, and the average value of the identification parameter corresponding to the low-quality content classification is greater than a fourth preset threshold.

[0194] According to one or more embodiments of the present disclosure, Example 7 provides the method of Example 6, wherein determining a target recognition result of the content to be recognized based on the plurality of recognition results further includes:

[0195] Obtaining recognition parameters corresponding to the original content categories respectively in the plurality of recognition results;

[0196] If the maximum value of the identification parameters of the multiple original content classifications is greater than the fifth preset threshold, and the identification parameters of the multiple similar content classifications do not meet the similarity identification condition, and the identification parameters of the multiple low-quality content classifications do not meet the low-quality identification condition, it is determined that the target identification result is an original content classification.

[0197] According to one or more embodiments of the present disclosure, Example 8 provides the method of any one of Examples 5-7, wherein the method further includes:

[0198] In the case where it is determined that the target recognition result is a similar content classification, determining a target position whose attention parameter in the fusion feature is greater than a preset threshold according to the fusion feature corresponding to the target content recognition pair, wherein the target content recognition pair is a content recognition pair corresponding to the maximum value of the recognition parameter of the similar content classification;

[0199] Determine the content corresponding to the target position in the candidate content of the target content recognition pair as the overlapping content corresponding to the content to be recognized;

[0200] The candidate content and the overlapping content in the target content identification pair are output.

[0201] According to one or more embodiments of the present disclosure, Example 9 provides the method of Example 1, wherein the dimension corresponds one-to-one to the feature extraction sub-model of the content recognition model, and the content recognition model is obtained in the following manner:

[0202] Acquire the training sample data, wherein the training sample data includes sample content, associated content corresponding to the sample content, and associated tags;

[0203] For each of the training sample data, according to the sample content and the associated content corresponding to the sample content, and the feature extraction sub-model corresponding to each dimension, determine the dimensional features corresponding to the sample content and the associated content in multiple dimensions respectively;

[0204] For each dimension, concatenate the features of the sample content and the associated content under the dimension to obtain a concatenated feature corresponding to the dimension;

[0205] Obtaining fusion features corresponding to the training sample data according to the splicing features corresponding to the multiple dimensions;

[0206] Determining the recognition result of the training sample data based on the classification sub-model of the content recognition model and the fusion feature;

[0207] According to the recognition result and the associated label, the parameters of the content recognition model are adjusted to obtain a trained content recognition model.

[0208] According to one or more embodiments of the present disclosure, Example 10 provides a content identification device, wherein the device includes:

[0209] A receiving module, used for receiving content to be identified;

[0210] a processing module, configured to obtain recognition results of the plurality of content recognition pairs according to the plurality of content recognition pairs including the content to be recognized and a content recognition model, wherein each of the content recognition pairs further includes a candidate content in a preset set, and the content recognition model is configured to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, wherein the multiple dimensions are used to represent multiple types of components in the content to be recognized;

[0211] The first determination module is used to determine a target recognition result of the content to be recognized based on the multiple recognition results.

[0212] According to one or more embodiments of the present disclosure, Example 11 provides a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processing device, the steps of the method described in any one of Examples 1-9 are implemented.

[0213] According to one or more embodiments of the present disclosure, Example 12 provides an electronic device, including:

[0214] a storage device having a computer program stored thereon;

[0215] A processing device is used to execute the computer program in the storage device to implement the steps of the method described in any one of Examples 1-9.

[0216] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0217] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0218] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims. Regarding the device in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

Claims

1. A content identification method, It is characterized in that The method comprises: Receiving content to be identified; According to a plurality of content recognition pairs including the content to be recognized and a content recognition model, recognition results of the plurality of content recognition pairs are obtained, wherein each of the content recognition pairs further includes a candidate content in a preset set, and the content recognition model is used to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, wherein the multiple dimensions are used to represent multiple types of components in the content to be recognized; Determining a target recognition result of the content to be recognized according to the recognition results of the plurality of content recognition pairs; The dimensions correspond one-to-one to the feature extraction sub-models of the content recognition model, and obtaining the recognition results of the multiple content recognition pairs according to the multiple content recognition pairs including the content to be recognized and the content recognition model includes: For each of the content identification pairs, determining dimensional features corresponding to the content to be identified and the candidate content in the content identification pair, respectively, in multiple dimensions according to the content to be identified and the candidate content, and the feature extraction sub-model corresponding to each dimension; For each dimension, concatenate the features of the content to be identified and the candidate content under the dimension to obtain a concatenated feature corresponding to the dimension; Obtaining fusion features corresponding to the content recognition pair according to the splicing features corresponding to the multiple dimensions; Based on the classification sub-model of the content recognition model and the fusion feature, a recognition result of the content recognition pair is determined.

2. The method according to claim 1, It is characterized in that The step of obtaining the fusion features corresponding to the content recognition pair according to the splicing features corresponding to the multiple dimensions includes: For each dimension, the splicing feature is processed based on the first attention layer to obtain an attention feature corresponding to the splicing feature; Concatenate the attention features in each dimension to obtain multi-dimensional concatenated features; The multi-dimensional splicing features are processed based on the second attention layer to obtain the fusion features.

3. The method according to claim 1, It is characterized in that The recognition result includes the recognition parameter of the content to be recognized corresponding to the classification, and the classification sub-model based on the content recognition model and the fusion feature determines the recognition result of the content recognition pair, including: According to the fusion feature and the classification sub-model, feature vectors of the fusion feature corresponding to a plurality of classifications are obtained, wherein the classifications include similar content classification, low-quality content classification, and original content classification; The feature vector is processed to obtain the recognition parameter of the fused feature corresponding to each category, so as to obtain the recognition result.

4. The method according to claim 1, It is characterized in that The recognition result includes the recognition parameter of the content to be recognized corresponding to the classification, and the determining the target recognition result of the content to be recognized according to the recognition results of the plurality of content recognition pairs includes: Obtaining recognition parameters corresponding to similar content categories in multiple recognition results; If the identification parameters of the plurality of similar content classifications satisfy similar identification conditions, the target identification result is determined to be a similar content classification, wherein the similar identification conditions include: The maximum value of the identification parameter corresponding to the similar content classification is greater than a first preset threshold; The maximum value of the identification parameters corresponding to the similar content classification is less than or equal to the first preset threshold, and the average value of the identification parameters corresponding to the similar content classification is greater than a second preset threshold.

5. The method according to claim 4, It is characterized in that The step of determining a target recognition result of the content to be recognized based on the recognition results of the plurality of content recognition pairs further includes: Obtaining recognition parameters corresponding to low-quality content categories respectively in the plurality of recognition results; If the identification parameters of the plurality of low-quality content classifications satisfy the low-quality identification condition, and the identification parameters of the plurality of similar content classifications do not satisfy the similar identification condition, it is determined that the target identification result is a low-quality content classification, wherein the low-quality identification condition includes: The maximum value of the identification parameter corresponding to the low-quality content classification is greater than a third preset threshold; The maximum value of the identification parameter corresponding to the low-quality content classification is less than or equal to the third preset threshold, and the average value of the identification parameter corresponding to the low-quality content classification is greater than a fourth preset threshold.

6. The method according to claim 5, It is characterized in that The step of determining a target recognition result of the content to be recognized based on the recognition results of the plurality of content recognition pairs further includes: Obtaining recognition parameters corresponding to the original content categories respectively in the plurality of recognition results; If the maximum value of the identification parameters of the multiple original content classifications is greater than the fifth preset threshold, and the identification parameters of the multiple similar content classifications do not meet the similarity identification condition, and the identification parameters of the multiple low-quality content classifications do not meet the low-quality identification condition, it is determined that the target identification result is an original content classification.

7. The method according to any one of claims 4 to 6, It is characterized in that The method further comprises: In the case where it is determined that the target recognition result is a similar content classification, determining a target position whose attention parameter in the fusion feature is greater than a preset threshold according to the fusion feature corresponding to the target content recognition pair, wherein the target content recognition pair is a content recognition pair corresponding to the maximum value of the recognition parameter of the similar content classification; Determine the content corresponding to the target position in the candidate content of the target content recognition pair as the overlapping content corresponding to the content to be recognized; The candidate content and the overlapping content in the target content identification pair are output.

8. The method according to claim 1, It is characterized in that The dimensions correspond one-to-one to the feature extraction sub-models of the content recognition model, and the content recognition model is obtained in the following manner: Acquire training sample data, wherein the training sample data includes sample content, associated content corresponding to the sample content, and associated tags; For each of the training sample data, according to the sample content and the associated content corresponding to the sample content, and the feature extraction sub-model corresponding to each dimension, determine the dimensional features corresponding to the sample content and the associated content in multiple dimensions respectively; For each dimension, concatenate the features of the sample content and the associated content under the dimension to obtain a concatenated feature corresponding to the dimension; Obtaining fusion features corresponding to the training sample data according to the splicing features corresponding to the multiple dimensions; Determining the recognition result of the training sample data based on the classification sub-model of the content recognition model and the fusion feature; According to the recognition result and the associated label, the parameters of the content recognition model are adjusted to obtain a trained content recognition model.

9. A content recognition device, It is characterized in that The device comprises: A receiving module, used for receiving content to be identified; a processing module, configured to obtain recognition results of the plurality of content recognition pairs according to the plurality of content recognition pairs including the content to be recognized and a content recognition model, wherein each of the content recognition pairs further includes a candidate content in a preset set, and the content recognition model is configured to obtain dimensional features of the content to be recognized and the candidate content in multiple dimensions, and determine the recognition result based on the multiple dimensional features, wherein the multiple dimensions are used to represent multiple types of components in the content to be recognized; A first determination module, configured to determine a target recognition result of the content to be recognized based on the recognition results of the plurality of content recognition pairs; The dimensions correspond one-to-one to the feature extraction sub-models of the content recognition model, and the processing module includes: A first determination submodule is used to determine, for each content recognition pair, dimensional features corresponding to the content to be recognized and the candidate content in multiple dimensions according to the content to be recognized and the candidate content in the content recognition pair, and the feature extraction submodel corresponding to each dimension; A first splicing submodule is used to splice the features of the content to be identified and the candidate content in each dimension to obtain a splicing feature corresponding to the dimension; A fusion submodule, used for obtaining fusion features corresponding to the content recognition pair according to the splicing features corresponding to the multiple dimensions; The second determination submodule is used to determine the recognition result of the content recognition pair based on the classification submodel of the content recognition model and the fusion feature.

10. A computer readable medium having a computer program stored thereon, It is characterized in that When the program is executed by a processing device, the steps of the method described in any one of claims 1 to 8 are implemented.

11. An electronic device, It is characterized in that include: a storage device having a computer program stored thereon; A processing device, configured to execute the computer program in the storage device to implement the steps of the method according to any one of claims 1 to 8.